cs.AI（2026-05-21）

📊 共 36 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (19 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (13 🔗1) 支柱八：物理动画 (Physics-based Animation) (3 🔗2) 支柱三：空间感知与语义 (Perception & Semantics) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (19 篇)

#题目一句话要点标签🔗⭐
1 Beyond Acoustic Emotion Recognition: Multimodal Pathos Analysis in Political Speech Using LLM-Based and Acoustic Emotion Models 利用LLM和声学情感模型进行政治演讲中的多模态情感分析，超越传统声学情感识别。 large language model multimodal
2 SciCore-Mol: Augmenting Large Language Models with Pluggable Molecular Cognition Modules SciCore-Mol：通过可插拔分子认知模块增强大型语言模型 large language model
3 Evaluating Large Language Models as Live Strategic Agents: Provider Performance, Hybrid Decomposition, and Operational Gaps in Timed Risk Play 在限时Risk游戏中评估大型语言模型作为实时战略智能体的性能 large language model
4 LLM-Metrics: Measuring Research Impact Through Large Language Model Memory 提出LLM-Metrics，利用大语言模型记忆评估研究影响力，无需引用数据。 large language model
5 A Camera-Cooperative ISAC Framework for Multimodal Non-Cooperative UAVs Sensing 提出相机协同的ISAC框架，用于多模态非合作无人机感知。 multimodal
6 Active Evidence-Seeking and Diagnostic Reasoning in Large Language Models for Clinical Decision Support 提出OSCE模拟器与诊断基准，揭示LLM在交互式临床诊断中证据搜寻的不足 large language model
7 Perception or Prejudice: Can MLLMs Go Beyond First Impressions of Personality? 提出GPR任务和MM-OCEAN数据集，揭示MLLM在人格感知中存在的偏见问题。 large language model multimodal
8 AtelierEval: Agentic Evaluation of Humans & LLMs as Text-to-Image Prompters 提出AtelierEval，用于评估人类和LLM作为文本到图像提示词生成器的能力。 large language model multimodal
9 LCGuard: Latent Communication Guard for Safe KV Sharing in Multi-Agent Systems 提出LCGuard，保障多智能体系统中基于KV缓存的隐式通信安全 large language model
10 AMEL: Accumulated Message Effects on LLM Judgments 揭示LLM评估中的累积消息效应(AMEL)，并提出缓解策略 large language model
11 Skill Weaving: Efficient LLM Improvement via Modular Skillpacks 提出SkillWeave框架以解决大语言模型多领域专门化问题 large language model
12 Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability 提出Synergistic Faithfulness以解决VLM可解释性问题 multimodal
13 Advancing Mathematics Research with AI-Driven Formal Proof Search 利用AI驱动的形式化证明搜索推进数学研究 large language model
14 Towards a General Intelligence and Interface for Wearable Health Data 提出可穿戴健康数据通用智能接口，通过大规模预训练实现个性化健康洞察。 foundation model
15 Meta-Soft: Leveraging Composable Meta-Tokens for Context-Preserving KV Cache Compression 提出Meta-Soft以解决KV缓存压缩中的信息损失问题 large language model
16 SGR-Bench: Benchmarking Search Agents on State-Gated Retrieval 提出SGR-Bench，用于评估智能体在状态门控检索任务中的表现 large language model ✅
17 IdleSpec: Exploiting Idle Time via Speculative Planning for LLM Agents IdleSpec：利用空闲时间进行推测性规划，提升LLM Agent性能 large language model
18 Not Yet: Humans Outperform LLMs in a Colonel Blotto Tournament Colonel Blotto博弈中，人类策略优于大型语言模型 large language model
19 Planning in the LLM Era: Building for Reliability and Efficiency 利用LLM生成可靠高效的规划器，解决智能体规划中的资源效率和可靠性问题 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (13 篇)

#题目一句话要点标签🔗⭐
20 Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning 提出Spreadsheet-RL，通过强化学习提升大语言模型在真实电子表格任务中的性能 reinforcement learning large language model
21 Deep Reinforcement Learning for Flexible Job Shop Scheduling with Random Job Arrivals 提出基于事件驱动深度强化学习的柔性作业车间调度方法，解决随机工件到达问题。 reinforcement learning deep reinforcement learning DRL
22 Efficient Agentic Reasoning Through Self-Regulated Simulative Planning 提出SR$^2$AM，通过自调节模拟规划实现高效的Agentic推理 reinforcement learning world model world models
23 Atom-level Protein Representation Learning Improves Protein Structure Prediction 提出TriProRep，通过原子级蛋白质表征学习提升蛋白质结构预测 representation learning VQ-VAE
24 Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention Gated DeltaNet-2：解耦线性注意力中的擦除与写入操作，提升长程依赖建模能力。 Mamba linear attention ✅
25 CLORE: Content-Level Optimization for Reasoning Efficiency CLORE：通过内容级优化提升大语言模型推理效率 reinforcement learning DPO large language model
26 LACO: Adaptive Latent Communication for Collaborative Driving LACO：一种自适应的潜在通信方法，用于提升协同驾驶性能 distillation foundation model
27 Dynamic Hypergraph Representation Learning for Multivariate Time Series without Prior Knowledge 提出DHACN模型，无需先验知识即可学习多元时间序列的动态超图表示，用于预测。 representation learning
28 Search-E1: Self-Distillation Drives Self-Evolution in Search-Augmented Reasoning Search-E1：通过自蒸馏驱动搜索增强推理中的自进化 distillation
29 SWE-Mutation: Can LLMs Generate Reliable Test Suites in Software Engineering? SWE-Mutation：评估LLM生成测试套件可靠性的基准与Agentic变异框架 reinforcement learning large language model
30 AI-Enabled Serious Games: Integrating Intelligence and Adaptivity in Training Systems 探讨AI赋能严肃游戏，实现智能化与自适应训练系统 reinforcement learning large language model
31 ACCoRD: Actor-Critic Conflict Resolution with Deep learning for O-RAN xApps 提出ACCoRD方法，利用深度强化学习解决O-RAN xApps中的冲突消解问题 reinforcement learning PPO
32 Unlocking Proactivity in Task-Oriented Dialogue 提出认知用户模拟器以解决主动任务导向对话问题 distillation reward shaping

🔬 支柱八：物理动画 (Physics-based Animation) (3 篇)

#题目一句话要点标签🔗⭐
33 Enhancing Visual Token Representations for Video Large Language Models via Training-Free Spatial-Temporal Pooling and Gridding 提出ST-GridPool，一种免训练的视觉token增强方法，提升视频大语言模型性能。 spatiotemporal large language model multimodal ✅
34 ST-SimDiff: Balancing Spatiotemporal Similarity and Difference for Efficient Video Understanding with MLLMs ST-SimDiff：平衡时空相似性和差异性，提升MLLM长视频理解效率 spatiotemporal large language model multimodal ✅
35 Characterizing the Fault Response of the Intel Neural Compute Stick 2 Under Single-Pulse Electromagnetic Fault Injection 电磁故障注入揭示NCS2在边缘AI应用中存在的严重可靠性问题 PULSE

🔬 支柱三：空间感知与语义 (Perception & Semantics) (1 篇)

#题目一句话要点标签🔗⭐
36 Knowledge Graph Re-engineering Along the Ontological Continuum (extended version) 提出本体连续体概念，用于知识图谱重构以适应神经符号AI需求 affordance

⬅️ 返回 cs.AI 首页 · 🏠 返回主页